A novel transformer‐embedded lithium‐ion battery model for joint estimation of state‐of‐charge and state‐of‐health

Abstract The state‐of‐charge (SOC) and state‐of‐health (SOH) of lithium‐ion batteries affect their operating performance and safety. The coupled SOC and SOH are difficult to estimate adaptively in multi‐temperatures and aging. This paper proposes a novel transformer‐embedded lithium‐ion battery model for joint estimation of state‐of‐charge and state‐of‐health. The battery model is formulated across temperatures and aging, which provides accurate feedback for unscented Kalman filter‐based SOC estimation and aging information. The open‐circuit voltages (OCVs) are corrected globally by the temporal convolutional network with accurate OCVs in time‐sliding windows. Arrhenius equation is combined with estimated SOH for temperature‐aging migration. A novel transformer model is introduced, which integrates multiscale attention with the transformer’s encoder to incorporate SOC‐voltage differential derived from battery model. This model simultaneously extracts local aging information from various sequences and aging channels using a self‐attention and depth‐separate convolution. By leveraging multi‐head attention, the model establishes information dependency relationships across different aging levels, enabling rapid and precise SOH estimation. Specifically, the root mean square error for SOC and SOH under conditions of 15 °C dynamic stress test and 25 °C constant current cycling was less than 0.9% and 0.8%, respectively. Notably, the proposed method exhibits excellent adaptability to varying temperature and aging conditions, accurately estimating SOC and SOH.